The present invention relates generally to the field of image analysis, and more particularly to template matching.
Image recognition products capture content from an image. Some image recognition products employ a template. Some of templates are trained to improve performance (trained templates). Using trained templates to capture content is sometimes also called zonal optical character recognition (OCR). Zonal OCR relies on a set of zones (fields), each of which can be visualized as a bounding rectangle on an image. A field can be represented as a set of pixel coordinates, which are used to retrieve the content from an image. Because various documents include similar fields having various placements, various trained templates are required to properly analyze each document. Each trained template must be manually created. A trained template can have ten or more fields and there can be hundreds of trained templates, depending on how a system is used. Manual entry of hundreds or thousands of fields is time consuming, repetitive, and error prone.